Task where AI models answer natural language questions about video content, requiring understanding of visual scenes and temporal dynamics.